Titel:
Sex as Gibbs Sampling
Speaker:
Chris Watkins
Datum:
26.06.2012, 11:30-12:30
Ort:
S2|02 / A213
Abstract:
The talk presents a view of genetic algorithms in which breeding can be viewed
as Gibbs sampling. A light-weight slice sampling method that converges to the
same equilibrium distribution will be compared to it.
Bio:
Chris Watkins is a Reader in Artficial...
[read more]
Titel:
Sex as Gibbs Sampling
Speaker:
Chris Watkins
Datum:
26.06.2012, 11:30-12:30
Ort:
S2|02 / A213
Abstract:
The talk presents a view of genetic algorithms in which breeding can be viewed
as Gibbs sampling. A light-weight slice sampling method that converges to the
same equilibrium distribution will be compared to it.
Bio:
Chris Watkins is a Reader in Artficial Intelligence at the Department of Computer
Science in Royal Holloway, University of London. He has worked in Reinforcement
Learning, Kernel Methods and String Kernels, Information-Theoretic Analysis of
Evolution, Epidemiology, and Communications Technology responses to Epidemics,
Visualisation, Estimation of the Risk of Investment Portfolios. His PhD thesis where
he introduced the algorithm Q-Learning is the single most cited thesis in Reinforcement
Learning.